Simulations

Running simulations (experiments) from yml config

It’s possible to make a yml file where we can describe simulation setup. Let’s put config into test1.yml

strategy: test.model.Strategy1  # strategy class

description: 
  - Description of this experiment
  - May be multiline

parameters:            # what startegy parameters to use in this run
  timeframe: "4h"
  parameter1: 123

data:                  # what data should be used for simulation
  ohlc(4h): (use_as_aux_loader:=loader("BINANCE.UM", "4h", source="mqdb::nebula"))

simulation:
  instruments: ["BINANCE.UM:BTCUSDT"]
  capital: 100000.0
  commissions: "vip0_usdt"
  start: "2021-01-01"
  stop: "2025-01-21"
  aux_data: use_as_aux_loader  # here we could use loader defined in data section !
  debug: ERROR

Now we can run it using qubx cli:

> qubx simulate test1.yml -o /backtests/tests

Result will be stored into /backtests/tests folder.

Instruments from a folder

Instruments that the configured instrument lookup does not have (for example US stocks read from Yahoo) can be taken from a folder of instrument JSON files for this one run:

> qubx simulate test1.yml --lookup file:configs/instruments -o /backtests/tests

• The folder path may be relative; it is resolved against the working directory.
• The option takes precedence over ~/.qubx/config.json and the QUBX_INSTRUMENT_LOOKUP__* environment variables, and variations of the run that execute in parallel use the same folder.
• qubx validate test1.yml --lookup file:<folder> accepts the same option; there it only checks that the folder exists.

Data read errors

If reading market data fails during a simulation (for example a network error from a remote store), the read is reopened from where it started, up to 4 attempts with waits of 1, 2 and 3 seconds. Rows already received are skipped, so each row arrives once. If the last attempt fails, the simulation stops with that error instead of continuing without data for some symbols.

Variations (hyperparameter optimization)

It’s possible to run variations for some parameters (kind of optimization preparation). For that it’s enough to add variate section:

strategy: test.model.Strategy1  # strategy class

description: 
  - Description of this experiment
  - May be multiline

parameters:            # what startegy parameters to use in this run
  timeframe: "4h"
  parameter1: 123
  parameter1: 333

variate:
  parameter1: [10, 20, 30, 40, 50, 50, 60, 70, 80, 90, 100]
  parameter2: [10, 20, 30, 40, 50, 50, 60, 70, 80, 90, 100]
  with:  # here we can apply constraints on parameters
    parameter1, parameter2: parameter2 > parameter1
    parameter2: parameter2 <= 90

data:                  # what data should be used for simulation
  ohlc(4h): (use_as_aux_loader:=loader("BINANCE.UM", "4h", source="mqdb::nebula"))

simulation:
  instruments: ["BINANCE.UM:BTCUSDT"]
  capital: 100000.0
  commissions: "vip0_usdt"
  start: "2021-01-01"
  stop: "2025-01-21"
  debug: ERROR

Run this using same command

> qubx simulate test1.yml -o /backtests/tests